A mixed multivariable monitoring method for casting production process
Through the core master analysis and support vector data description method, an offline monitoring model for the casting production process was established, which solved the problem that univariate monitoring could not identify multiple performance indicators and process variables in complex processes, and realized the multivariate of the casting production process. Monitoring improves the accuracy of abnormal identification and data smoothness.
Patent Information
- Application Number
- CN202111424681.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-11-26
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2041-11-26
AI Technical Summary
During casting production, univariate monitoring cannot effectively identify the multi-performance indicators and process variables with high correlation and nonlinear characteristics in complex processes, making it difficult to control product quality fluctuations.
The kernel master analysis and support vector data description method are adopted to establish an offline monitoring model for the casting production process through data dimensionality reduction and super-spherical model, and the kernel matrix and feature vectors are used for data mapping and dimensionality reduction, and data pre-pre-data is pre-produced by exponential weighted moving average method. Processing, simplify monitoring process and identify abnormal states.
It realizes accurate monitoring of multivariables in the casting production process, reduces dimensions, improves the accuracy of abnormal identification and data smoothness, and is suitable for linear and nonlinear process monitoring.
Smart Images

Figure CN114167826B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of casting production process state monitoring, and in particular to a mixed multi-variable monitoring method for casting production process. Background Art
[0002] The foundry industry is an important part of modern advanced manufacturing, and promoting intelligent foundry is also important for the development of manufacturing.
[0003] Investment casting has many complex production processes and a long production cycle. There are many factors that affect the final quality of castings. Factors such as people, machines, materials, methods, measurements and environment in the production process will disturb the quality of the process, resulting in fluctuations in product quality. At the same time, customers have higher and higher requirements for casting quality. Therefore, investment casting quality management is one of the problems faced by enterprises. In order to improve process monitoring capabilities and product quality, foundry companies have used traditional statistical process control. On the basis of the 6σ principle, Shewhart control charts, CUMUS control charts, EWMA control charts, etc. are added to monitor some important indicator variables in the production process separately. The monitoring of single variables has a good effect and improves product quality to a certain extent. Due to the complexity of investment casting processes and influencing factors, it is necessary to monitor multiple performance indicators and process variables of the product. There is a high correlation between these process variables. The correlation of the process variables themselves with time series cannot be accurately identified by using only single variable statistical process control.
[0004] There are many uncertain factors that affect the quality of complex investment castings such as aerospace. In order to study the fluctuation of product quality, enterprises must strengthen the monitoring and control of quality management in the production process and realize the online monitoring of important process parameters of each process in the production process of complex investment castings. The current process variables have mixed variable correlation. How to accurately extract abnormal information that can reflect the advantages of process operation status and effectively utilize it based on the nonlinear, non-Gaussian and strong coupling characteristics of process data has become an important issue in the monitoring of mixed multi-variable casting production processes. Summary of the invention
[0005] In view of the defects of the prior art, the purpose of the present invention is to provide a mixed multi-variable monitoring method for the casting production process, aiming to solve the problem of single variable monitoring in the casting process and inability to monitor multiple performance indicators and process variables with high correlation in the casting process.
[0006] To solve the above problems, according to one aspect of the present invention, a method for establishing a casting production offline monitoring model is provided, comprising the following steps:
[0007] (1) Data collection step: collecting historical data of the normal operation of the casting process as sample data, wherein the sample data includes sensor data and non-sensor data, and preprocessing the sample data to obtain processed data;
[0008] (2) Data dimensionality reduction step: Use the kernel principal component analysis method to reduce the dimensionality of the processed data, map the processed data from its original data space to the feature space through the Gaussian kernel function, and calculate the kernel matrix. Use the centralization method to correct the kernel matrix to obtain the corrected kernel matrix, calculate the eigenvector of the kernel matrix, and use the Schmidt orthogonalization method to orthogonalize the unitized eigenvector. Use the cumulative contribution rate method to obtain the principal component whose contribution rate is greater than the preset value, and calculate the score matrix in the kernel matrix where the principal component is located, that is, the principal component space;
[0009] (3) Data description step: Support vector data description (SVDD) is used to perform a hypersphere description on the score matrix to obtain a hypersphere model, in which the radius of the hypersphere is used as the judgment statistic.
[0010] Furthermore, the preprocessing in the data collection step (1) includes centralization processing and dimensionless processing.
[0011] Furthermore, the preprocessing also includes using an exponentially weighted moving average method to perform noise reduction on the sensor data to remove the time series correlation of the sensor data.
[0012] Furthermore, the sample data in the data acquisition step (1) has complex characteristics, including nonlinearity, non-Gaussianity, and multiple operating conditions.
[0013] According to another aspect of the present invention, a hybrid multivariable monitoring method based on the above model is provided for a casting production process, comprising the following steps:
[0014] (1) Sample monitoring step: collecting real-time data of the casting process, wherein the real-time data includes sensor data and non-sensor data, and pre-processing the real-time data to obtain monitoring samples;
[0015] (2) Sample feature calculation steps: The kernel principal component analysis method is used to reduce the dimension of the monitoring samples. The monitoring samples are mapped from their original data space to the feature space through the Gaussian kernel function, and the real-time kernel matrix is calculated. The real-time kernel matrix is corrected by the centralization method to obtain the real-time corrected kernel matrix. The real-time corrected kernel matrix is multiplied with the unitized feature vector in the foundry production offline monitoring model to obtain the real-time score matrix.
[0016] (3) Sample description step: Use SVDD to perform a hyperspherical description of the score matrix and obtain the distance between the observed data points of the real-time data and the center of the hypersphere as the monitoring statistic;
[0017] (4) Comparison and judgment step: If the monitoring statistic is less than the judgment statistic, it means that the casting process is in a normal state, otherwise the casting process is in an abnormal state.
[0018] Furthermore, the preprocessing in the sample monitoring step (1) includes centralization processing and dimensionless processing.
[0019] Furthermore, the preprocessing also includes using an exponentially weighted moving average method to perform noise reduction on the sensor data to remove the time series correlation of the sensor data.
[0020] Furthermore, the real-time data in the sample monitoring step of step (1) has complex characteristics, including nonlinearity, non-Gaussianity and multiple working conditions.
[0021] In general, the technical solution of the present invention is used to achieve the following beneficial effects compared with the prior art:
[0022] 1. The support vector data description algorithm adopted in the present invention is based on an unsupervised learning algorithm, which simplifies multivariable monitoring into monitoring the distance from the center of a hypersphere, thus simplifying the monitoring process;
[0023] 2. Compared with single variable monitoring, the present invention considers the correlation between mixed multivariables, reduces the dimension of data, and can more accurately identify possible anomalies in the production process;
[0024] 3. The present invention uses the exponentially weighted moving average method to smooth the data, reduce the impact of time series on the data, and improve the accuracy of the data;
[0025] 4. The present invention introduces the kernel method into principal component analysis to extract the nonlinear characteristics of the process, which can effectively solve the nonlinear problem, so that the present method can be applied to both linear process monitoring and nonlinear process monitoring. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 is a real-time casting process monitoring flow chart constructed according to a preferred embodiment of the present invention;
[0027] Figure 2 is a flow chart of kernel principal component analysis constructed according to a preferred embodiment of the present invention;
[0028] Figure 3 It is a rendering of the SVDD method constructed according to the preferred embodiment of the present invention. DETAILED DESCRIPTION
[0029] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0030] like Figure 1 As shown, the present invention proposes a mixed multivariable monitoring method for the casting production process in view of the multivariable, nonlinear, strongly coupled and high-dimensional characteristics of the on-site production process data of the casting enterprise. The monitoring method for the mixed multivariable is generally divided into two links. First, an offline monitoring model is established according to the normal operation process data, and then the established model is used for on-site monitoring. The method includes the following steps:
[0031] S1. Data preparation steps: Select important processes in the investment casting process, such as the pouring process of the intermediate casing, and select the corresponding data table (intermediate casing pouring record table) from the historical database. The 9 dimensions of key process parameters that affect the process include shell temperature, pouring temperature, leakage rate, vacuum degree, voltage, current, cooling water pressure, cooling water inlet temperature and cooling water outlet temperature. It can be seen that there are characteristics of mixed multivariable multi-dimensional high coupling; select 100 groups of normal data as sample data for training, and get the initial data X∈R 100x9 ;
[0032] S2. Data preprocessing step: The pouring temperature is between 300 and 400℃, while the pressure is only a few tenths. The pouring temperature will obliterate the data with small absolute values in the modeling process, so data standardization is a very important link, which can highlight the correlation of process variables to a certain extent and eliminate the influence of dimension on sample data.
[0033] Data preprocessing consists of two sub-steps:
[0034] S21 Data centralization: Data centralization is to translate the data, subtract the mean of the data in this dimension from each data, so that the data coincides with the center of the sample set;
[0035] S22 dimensionless processing, dimensionless processing of data is to use variance normalization, divide each column of data by its respective standard deviation, so that each variable has equal weight;
[0036] S3. Voltage and current are easily affected by equipment vibration noise, and there is also sensor measurement noise. The sampling results of the cooling water inlet and outlet temperatures are affected by the sampling values at the previous moment. Therefore, the exponentially weighted moving average method is used to smooth the data, enhance the correlation between variables, and reduce the influence of the data observation values at the previous and next moments. It also has a certain ability to remove noise. The key parameter weighting factor of the exponentially weighted moving average method is 0.5. The farther the sample point is from the current one, the smaller the weight.
[0037] S4. The sample data has high dimensionality, such as Figure 2 As shown, principal component analysis is used to reduce the dimension of the data, and the kernel method is introduced to calculate the kernel matrix K. The data in the original data space is mapped to the feature space through the kernel function. The kernel function used is the Gaussian kernel function, and the penalty coefficient of the kernel function is c=800. The centralized kernel matrix Kc is used to correct the kernel matrix, calculate the eigenvalues and eigenvectors V of the matrix Kc, and arrange the eigenvalues and eigenvectors from large to small. The larger the eigenvalue, the more useful information it contains. The Schmidt orthogonalization method is used to orthogonalize and normalize the eigenvector A, and the cumulative contribution rate method is used to obtain 3 principal component scores greater than the preset value (85% in this embodiment). The corresponding eigenvalues of the first 3 principal components are 6.18, 1.78, and 1.49, respectively. In order to finally reduce the 9 parameter variables in the casting process to 3 dimensions, the 3 principal component scores are used to replace the original 9-dimensional data information, and the score matrix T is calculated in the principal component space;
[0038] S5. Input the score matrix T into SVDD and obtain the following after training: Figure 3 The hypersphere shown has a center a and a radius R, which makes this hypersphere the smallest hypersphere and can contain most normal samples. For the casting process, the radius of the hypersphere is calculated to be R=38.34, which is used as a statistic for process monitoring.
[0039] S6. The key process parameters of the corresponding process are collected in real time through the data collection system, and the parameters and methods of the training samples (i.e., S2, S3, and S4) are used as the real-time monitoring data to process the data and obtain the score matrix of the monitoring data;
[0040] S7. Use the score matrix of S6 as input, and use the hypersphere model trained in S5 to calculate the distance d from the monitoring data point to the center of the hypersphere. Use the distance d to measure the degree of abnormality of the monitoring data. If the distance d is less than the radius R of the hypersphere, that is, it is within the hypersphere, it means that the data is in a normal state. Otherwise, the data is in an abnormal state and the monitoring data has shifted. When an abnormal situation occurs, the operator should stop the current process in time and continue to run after adjusting the production process conditions.
[0041] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for establishing an offline monitoring model for casting production, characterized in that: The steps include: (1) Data collection step: collecting historical data of the normal operation of the casting process as sample data, wherein the sample data includes sensor data and non-sensor data, and preprocessing the sample data to obtain processed data, wherein the preprocessing includes using an exponentially weighted moving average method to perform noise reduction on the sensor data; (2) Data dimensionality reduction step: Use the kernel principal component analysis method to reduce the dimensionality of the processed data. Use the Gaussian kernel function to map the processed data from its original data space to the feature space, and calculate the kernel matrix. Use the centralization method to correct the kernel matrix to obtain the corrected kernel matrix. Calculate the eigenvector of the kernel matrix. Use the Schmidt orthogonalization method to obtain the unitized eigenvector. Use the cumulative contribution rate method to obtain the principal components whose contribution rate is greater than the preset value. Calculate the score matrix in the kernel matrix where the principal components are located, i.e., the principal component space. (3) Data description step: SVDD is used to perform a hypersphere description on the score matrix to obtain a hypersphere model, in which the radius of the hypersphere is used as the judgment statistic.
2. The method according to claim 1, characterized in that The preprocessing in the data collection step (1) includes centralization processing and dimensionless processing.
3. The method according to claim 1, characterized in that The sample data in the data acquisition step (1) has complex characteristics, including nonlinearity, non-Gaussianity and multiple working conditions.
4. A mixed multivariable monitoring method for a casting production process, wherein the casting production process is monitored using a hypersphere model established by the method for establishing a casting production offline monitoring model as described in any one of claims 1 to 3, wherein: The steps include: (1) Sample monitoring step: collecting real-time data of the casting process, the real-time data including sensor data and non-sensor data, pre-processing the real-time data to obtain monitoring samples; (2) Sample feature calculation step: using a kernel principal component analysis method to perform dimensionality reduction processing on the monitoring samples, mapping the monitoring samples from their original data space to the feature space through a Gaussian kernel function, and calculating a real-time kernel matrix, using a centralization method to correct the real-time kernel matrix to obtain a real-time corrected kernel matrix, and multiplying the real-time corrected kernel matrix with the unitized feature vector in the casting production offline monitoring model to obtain a real-time score matrix; (3) Sample description step: Use SVDD to perform a hypersphere description on the real-time score matrix, and obtain the distance between the observed data points of the real-time data and the center of the hypersphere as the monitoring statistic; (4) Comparison and judgment step: If the monitoring statistic is smaller than the judgment statistic, it means that the casting process is in a normal state; otherwise, the casting process is in an abnormal state.
5. The method according to claim 4, characterized in that The preprocessing in the sample monitoring step (1) includes centralization processing and dimensionless processing.
6. The method according to claim 5, characterized in that The preprocessing also includes using an exponentially weighted moving average method to perform noise reduction processing on the sensor data.
7. The method according to claim 4, characterized in that The real-time data in the sample monitoring step of step (1) has complex characteristics, including nonlinearity, non-Gaussianity and multiple working conditions.
Citation Information
Cited By
Real-time monitoring system, method and equipment for aluminum alloy plate casting process
CN121254714A
An aluminum alloy plate casting process real-time monitoring system, method and device
CN121254714B
A digital processing method for monitoring and optimizing an investment casting process
CN122755521A
A digital processing method for monitoring and optimizing an investment casting process
CN122755521B
